1 / 8

Cliffs, Spikes and Valleys

Cliffs, Spikes and Valleys. Goetz Graefe and Harumi Kuno. 50% of power is load-independent. Clearly we need workload optimization: Increase server utilization Decrease power wasted by idle machines Without sacrificing performance.  So what’s the problem?

Télécharger la présentation

Cliffs, Spikes and Valleys

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Cliffs, Spikes and Valleys Goetz Graefe and Harumi Kuno

  2. 50% of power is load-independent • Clearly we need workload optimization: • Increase server utilization • Decrease power wasted by idle machines • Without sacrificing performance.  • So what’s the problem? • Hard to predict cliffs, spikes, and valleys in mixed workloads • Hard to plan  Safer to over-provision

  3. Scan one index, apply 2nd predicate Example: System B has a cliff System A System B

  4. Example: contention moves cliffs overflow

  5. Example valley and spike: online index creation Load Combined load System capacity Query load Indexing load Time SMDB 2011 - Hannover

  6. For example: prevent a spike by accommodating a valley Load System capacity Query load Delayed indexcompletion Time SMDB 2011 - Hannover

  7. How to motivate, evaluate, protect improvements in robust query processing? • Metrics for “good performance every time” • As opposed to optimal performance, predictability or progress estimation. • Metric that rewards lack of cliffs, spikes, and valleys, • Not just predict them • Improvements include: • Robust algorithms (e.g., memory-adaptive sorting) • Robust algorithm choices (e.g., g-join and g-distinct) • Robust plans (e.g., dynamic re-ordering of operations within a plan), etc.

  8. Nutshell • How can (data management) software prevent over-provisioning? • Workload management and optimization : great, but what about the cliffs, spikes, and valleys? • What are good metrics and tests for “good performance every time” ? • How can we improve query processing robustness?

More Related